Gianluca Guida <mailto:gianluca.guida@xxxxxxxxxxxxx> wrote:
> Jiang, Yunhong wrote:
>> Thanks for your really quick-hand implementation. I will
> update my another patch accordingly tomorrow.
>>
>> So still one question to the two assertion (or to Tim??) in
> sh_rm_write_access_from_sl1p()/sh_put_ref(). What's the
> potential error to be protected by this checking? If page is a
> shadow, it's count_info will always be 0, right? Or it is just
> a sanity checking?
>
> They're sanity checking. Checks are there to be sure that the
> page is a
> shadow page. It used to be "->mbz == 0", which was clearer, in the old
> page_info model.
>
> In the sh_rm_write_access_from_sl1p(), the check is there to
> ensure that
> the page is *still* a shadow page. As for current code, though, that check
> should be useless.
>
> I actually think that having a more precise way of knowing when a page
> is a shadow page is definitely needed, for debugging or even
> in case the
> shadow allocation becomes more dynamic in the future.
I think a shadow page has always count_info == 0, while count_info==0 does not
always mean a shadow pages.
>
>> I need change this is because, if we mark a page offline,
> then the count_info is not 0, even for shadow page. Can I just checking the
> count_mask here?
>
> Ehr, I am not following the discussions about your patch, but
> I suppose
> you set the page off-line after all references and uses of the page,
> even inside shadows, are removed. Or the count_info is != 0 to signal
> that the page is going offline?
The latter. We will mark the page offline pending through a new flag. When we
free the page, we will check the flag to decide the action. So the checking is
shadow is not valid anymore. I will update it according to Keir's feedback.
Thanks
Yunhong Jiang
>
> Thanks,
> Gianluca
_______________________________________________
Xen-devel mailing list
Xen-devel@xxxxxxxxxxxxxxxxxxx
http://lists.xensource.com/xen-devel
|
Home